HYBRID TRANSFORMERS AND CLASSIFIERS FOR MALICIOUS URL DETECTION

Author

Sukeerthi Sri Ram, Southern Illinois University CarbondaleFollow

Date of Award

5-1-2025

Degree Name

Master of Science

Department

Electrical and Computer Engineering

First Advisor

Weng, Ning

Abstract

In the fast-growing era of cybersecurity effectively identifying malicious URLs is very important to protecting users from the threats like phishing and malware. Traditional URL detection methods often rely on manually generated features and heuristic approaches which are increasingly inadequate against sophisticated cyberattacks. For addressing the challenges faced by the traditional models our research introduces a new model that combines the strengths of two advanced transformer-based language models called BERT (Bidirectional Encoder Representations from Transformers) and BART (Bidirectional and Auto-Regressive Transformers). BERT is proficient at understanding the context of words within a given sentence using its bidirectional training and capturing the relationships in the data. data. BART excels in the sequence-sequence tasks very effectively and helps in reconstructing the corrupted text sequences which is very important in understanding the sequential nature of the URL. Our approach aims to enhance the accuracy and robustness of the malicious URL detection system. We conducted extensive experiments on a diverse dataset comprising both malicious and benign URLs to evaluate the efficacy of our hybrid model. The implications of this research are substantial. By harnessing the capabilities of BERT and BART, cybersecurity systems can more effectively identify and mitigate threats posed by malicious URLs, thereby providing enhanced protection for users against evolving cyber threats. This study underscores the importance of adopting advanced machine learning models in the ongoing battle against cybercrime.